Classical conditioning with pulsed integrated neural networks:circuits and system

摘要

In this paper, we investigate on-chip learning for pulsed,nintegrated neural networks. We discuss the implementational problems thentechnology imposes on learning systems, and we find that a biologicallyninspired approach using simple circuit structures is most likely tonbring success. We develop a suitable learning algorithm-ancontinuous-time version of a temporal differential Hebbian learningnalgorithm for pulsed neural systems with nonlinear synapses-as well asncircuits for the electronic implementation. Measurements from annexperimental CMOS chip are presented. Finally, we use our test chip tonsolve simple classical conditioning tasks, thus verifying the designnmethodologies put forward in the paper